Welcome
to the
Classification of Watertight Models Track
of SHape Retrieval Contest 2008

Track Goal:

This track addresses 3D shape classification, meant as the process of assigning an unknown 3D model to one of the classes of a given set of models. The latter will be pre-classified according to semantic criteria, such as functionality (e.g., "objects for drinking") or presence of characteristic shape features (e.g., "parts with sharp features"). Different levels of categorization will be taken into account, ranging from an entry, coarse level (e.g., grouping objects into "furniture" and "four limbs animals") to finer levels (e.g., "chairs", "sofas", "dogs", and "cats"). The algorithms proposed by the track competitors should be able to capture the geometric and semantic meaning stored in each category. This can be achieved through a learning phase, as specified below.

Track Description:

First off, the participants will be provided with a small Training Set (TS), pre-classified as described before, that can possibly be used for training the algorithms in a short-term learning phase. Then, a larger, pre-classified Data Set (DS) will be released, along with a Query Set (QS) of unknown objects. For each query in QS, the aim is to establish a ranking of memberships in the DS classes, where the class ranked first represents the category that most reasonably contains the query. Hence, the participants will be asked to return, for each query, a list of values corresponding to the probability for the query to belong to each of the DS classes. All models in TS, DS and QS will be watertight models. The taxonomy used for classifying the TS, DS and QS will be innerly coherent and established a priori. The performance of the algorithms will be evaluated using standard parameters (e.g., the classification rate).

Instructions for competitors:

Each participant is requested to:

• Register to the track by sending an email to both Daniela Giorgi (email: daniela.giorgiATge.imati.cnr.it) and Simone Marini (email: simone.mariniATge.imati.cnr.it), with "Registration for SHREC08: Classification Track" as the subject.
• Download from our ftp server the TS (available from the 10th of March 2008), and the DS and QS (available from the 12th of March 2008). All 3D models are represented via OFF files. Participants receive login info via email.
• Provide by the 14th of March 2008:
  • A plain ASCII file for each membership matrix. Entry (i,j) of the matrix is the probability that query i of QS belongs to class j of DS. Probability values should range in the interval [0,1] and sum to 1. For each categorization level, up to 3 membership matrices may be submitted, resulting from different algorithms or parameter settings.
  • The executables that generated the results submitted. We accept Windows, Linux or Mac executables (scripts are also acceptable), that should be provided along with all the possible dependencies (.dll, .so, other executables...). Alternatively, the source code may be provided, together with full compilation instructions and their dependencies: we accept C, C++ and Matlab code.
  • A paper (up to 2 pages) including a summary of the classification technique and performances, along with a comparison of the results against the others in the track. Please provide a first version of this paper, presenting the description of the classification technique, and leave some space for the section on results/discussions, as this part can only be filled after all the results are in. The deadline for submitting the final version of the paper is the 24th of March 2008. See below for styling requirements.

Track proceedings:

This year the summary papers will be included in the SMI 2008 proceedings, so we require the participants to respect the following styling rules:

• The title of the short paper must start with "SHREC'08 Entry: [your title]".
• Please, follow these SMI 2008 paper submission guidelines carefully;
• Use grayscale images only. XnView can easily convert RGB images to (256 scale) Gray images. Note that RGB images with R=G=B may suffer a "color-shift" when printed using the CMYK color model (which will be the case).
• Send the sources of your short paper (Word or Latex), images, etc.

Notes:

Within this track, a model is considered "watertight" if it is represented by seamless surfaces (without defective holes or gaps).